Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Structured, State-Aware, and Execution-Grounded Reasoning for Software Engineering Agents

Tse-Hsun, Chen

TL;DR

The paper identifies a fundamental mismatch between current reactive SE agents and the needs of long-horizon software engineering tasks, highlighting how lack of explicit structure and persistent state impedes coherent reasoning. It proposes a shift toward structured, state-aware, and execution-grounded reasoning, emphasizing explicit intermediate representations, evolving internal state, and integrated execution feedback. The authors outline a concrete roadmap—adopting finite-state-like representations with pre-/post-conditions, treating reasoning as an evolving state, and mapping feedback to structured updates—to enable more reliable multi-step reasoning. This approach aims to yield more coherent, state-consistent SE agents capable of handling real-world, long-duration development tasks.

Abstract

Software Engineering (SE) agents have shown promising abilities in supporting various SE tasks. Current SE agents remain fundamentally reactive, making decisions mainly based on conversation history and the most recent response. However, this reactive design provides no explicit structure or persistent state within the agent's memory, making long-horizon reasoning challenging. As a result, SE agents struggle to maintain a coherent understanding across reasoning steps, adapt their hypotheses as new evidence emerges, or incorporate execution feedback into the mental reasoning model of the system state. In this position paper, we argue that, to further advance SE agents, we need to move beyond reactive behavior toward a structured, state-aware, and execution-grounded reasoning. We outline how explicit structure, persistent and evolving state, and the integration of execution-grounded feedback can help SE agents perform more coherent and reliable reasoning in long-horizon tasks. We also provide an initial roadmap for developing next-generation SE agents that can more effectively perform real-world tasks.

Towards Structured, State-Aware, and Execution-Grounded Reasoning for Software Engineering Agents

TL;DR

The paper identifies a fundamental mismatch between current reactive SE agents and the needs of long-horizon software engineering tasks, highlighting how lack of explicit structure and persistent state impedes coherent reasoning. It proposes a shift toward structured, state-aware, and execution-grounded reasoning, emphasizing explicit intermediate representations, evolving internal state, and integrated execution feedback. The authors outline a concrete roadmap—adopting finite-state-like representations with pre-/post-conditions, treating reasoning as an evolving state, and mapping feedback to structured updates—to enable more reliable multi-step reasoning. This approach aims to yield more coherent, state-consistent SE agents capable of handling real-world, long-duration development tasks.

Abstract

Software Engineering (SE) agents have shown promising abilities in supporting various SE tasks. Current SE agents remain fundamentally reactive, making decisions mainly based on conversation history and the most recent response. However, this reactive design provides no explicit structure or persistent state within the agent's memory, making long-horizon reasoning challenging. As a result, SE agents struggle to maintain a coherent understanding across reasoning steps, adapt their hypotheses as new evidence emerges, or incorporate execution feedback into the mental reasoning model of the system state. In this position paper, we argue that, to further advance SE agents, we need to move beyond reactive behavior toward a structured, state-aware, and execution-grounded reasoning. We outline how explicit structure, persistent and evolving state, and the integration of execution-grounded feedback can help SE agents perform more coherent and reliable reasoning in long-horizon tasks. We also provide an initial roadmap for developing next-generation SE agents that can more effectively perform real-world tasks.
Paper Structure (13 sections, 1 figure)

This paper contains 13 sections, 1 figure.

Table of Contents

  1. Introduction
  2. The Fundamental Mismatch in SE Agent Reasoning
  3. Why SE Tasks Expose the Limits of Reactive Agents
  4. How the Reactive Paradigm Fails in Practice
  5. Inconsistent reasoning reconstructions.
  6. Forgotten assumptions and broken reasoning process.
  7. Interpreting execution feedback in isolation.
  8. Retrying without knowing where to return.
  9. Rethinking SE Agent Reasoning: From Reactive to Structured
  10. From Conversation Histories to Explicit Intermediate Representations.
  11. Reasoning as an Evolving State Rather Than Repeated Reconstruction
  12. Connecting Execution Feedback to Structured Updates
  13. Conclusion

Figures (1)

  • Figure 1: An illustration of the difference between reactive SE agents and structured, state-aware, execution-grounded agents. The example highlights the explicit persistence and evolution of agent state across actions.